Developing Alternative Multilinear Regression-Based Intelligence Hybrid Model

The study focuses on improving quantitative structure–property relationships (QSPR) simulation for mycotoxins by employing multi-oriented data-intelligent approaches. While previous methods only considered the molecular structure, this study aims to overcome this limitation using four data-driven model approaches based on mycotoxins’ retention time (tR). The goal is to leverage the correlation between various molecular properties and develop a hybrid intelligence-based model. The three hybrid intelligence models examined in this study are MLR-SVM, MLR-ANN, and MLR-ANFIS. Four molecular properties, namely retention index, peak symmetry, mono isotopic mass, and relative sensitivity factor, are selected as input variables for modelling. The relationships between the input and output parameters are measured using Spearman Pearson correlation. Several metrics are employed to evaluate the fitness, performance, and adequacy of the models. These include the coefficient of determination, root mean square error, mean square error and correlation coefficient. The results indicate that the ANFIS model outperformed the other three single models (MLR, SVM, and ANN). Furthermore, the hybrid models outperformed the single models, with the hybrid model MLR-ANFIS being the adequate hybrid model. Based on these findings, the hybrid MLR-ANFIS intelligence model could be an alternative high-performance model for simulating QSPR in mycotoxins. This highlights the potential of using hybrid intelligence approaches that combine multiple techniques to enhance the accuracy and effectiveness of QSPR simulations.